About Avencera

How it works

Avencera runs open models, such as NVIDIA Parakeet for transcription and pyannote’s community-1 for speaker labels, behind OpenAI’s transcription API. You buy prepaid credit, by card or with bitcoin over Lightning, and pay only for the audio you send.

We never train on your audio. A direct request’s audio is deleted once its transcript is delivered, and the deletion is retried if it fails. A job’s audio is kept for 24 hours at most.

Open source

The models run on open-source Rust libraries that Avencera publishes:

• speakrs, for speaker diarization
• scriptrs, for transcription

Who writes here

Praveen Perera writes the engineering notes and the comparisons, and maintains the libraries above. Find more at praveenperera.com.